JP2011072843A - Tachyarrhythmia detection and discrimination based on curvature parameter - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To estimate a frequency of a sampled cardiac rhtym signal and to classify the rhythm. <P>SOLUTION: The received signal is sampled and transformed into a curvature series. A lobe in the curvature series corresponds to a characteristic point in the sampled series. Characteristic points are selected based on a time of a lobe in the curvature series and, in one embodiment, an amplitude of the signal at the time of the lobe. A frequency of the sampled series is estimated by autocorrelating a function of the series of the characteristic points. In one embodiment, the function is a time difference function. The rhythm is classified by plotting the timewise proximity of characteristic points derived from an atrial signal with characteristic points derived from a ventricular signal. Regions of the plot are associated with a particular rhythm, and the grouping of the data corresponds to the classification. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates generally to heart rate rhythm management, and more particularly, but not exclusively, to arrhythmia detection and identification in implantable medical devices.

  This application is filed October 31, 2000, which is incorporated herein by reference in its entirety, by inventor Sweeney et al., “CURVATURE BASED METHOD THE FOR SELECTING FROM AN ELECTROLOGY. SIGNALS FOR PURPOSE OF COMPLEX IDENTIFICATION AND CLASSIFICATION related to co-pending and co-pending US patent application Ser. No. 09 / 703,269.

  This application is filed on July 15, 2002, the specification of which is incorporated herein by reference in its entirety, as “USE OF CURVATURE BASED FEATURES FOR BEAT DETECTION” by inventor Sweeney. And related to co-pending US patent application Ser. No. 10 / 195,838 of the same assignee.

  This application is entitled “SIGNAL COMPRESSION BASED ON CURVATURE PARAMETERS”, filed on June 27, 2003, which is incorporated herein by reference in its entirety. Related to co-pending US patent application Ser. No. 10 / 606,820, to the same assignee.

  Implantable heart rate rhythm management devices generally treat heartbeat signals with a monitor to treat the heart. For treatment, pacing pulses are sent to trigger heart contraction or to apply a defibrillation impact to prevent heart rhythm abnormalities. Some heart rate rhythm management devices also monitor cardiac function or other physiological parameters for use in pacing pulse delivery control.

  Some implantable heart rate rhythm management devices are programmed to verify that an arrhythmia has occurred prior to treatment. Furthermore, if the rhythm type is known, it is possible to adjust the treatment according to the needs of the patient.

  It is an object of the present invention to provide an improved arrhythmia detection and identification system, apparatus, and method that allows for more accurate diagnosis and patient treatment.

  In accordance with the present invention, a curvature-based method for selecting features from a sample signal is provided. In one embodiment, the sample signal includes a heartbeat signal or an electrocardiogram. According to one aspect of the present invention, the characteristic points are selected by detecting the heartbeat signal and calculating the curvature continuously for each sample received in real time. Characteristic points are sometimes called singularities or data points.

  In one embodiment, each “turn” of the signal is represented by a characteristic point having time, value, area, and width criteria. Time represents the time indicating the center of the turn in a continuous curve. The value represents the amplitude of the input signal at the center time of the turn. The area represents a value indicating the direction and degree of the turn. The width represents the period during which the turn occurs.

  According to the method presented herein, each new characteristic point becomes apparent upon completion of the turn it draws. In one embodiment, the series of characteristic points is analyzed sequentially as each new characteristic point occurs. In one embodiment, the characteristic points are stored in a buffer for later analysis.

  Not all turns in the input signal are the same. Some turns are sharply deflected at an acute angle (eg, turns in a QRS signal), while others are small and do not lead to large deflections (eg, noise in the signal). In one embodiment, the present invention includes a scheme that selects significant characteristic points of a signal and discards smaller characteristic points or only a few signal deflections that may represent noise.

  The area of the characteristic point is an indirect measure of the angular turn of the signal. The maximum turn is about 180 degrees which represents a very rapid signal reversal. A large deflection in the signal, such as an R wave (a complete signal oscillation in 10 samples over a period of 50 ms) will have a large turn angle of 178 degrees. A small noise spike (such as a 10% full-scale vibration of the signal sample) can also have a large turn angle of 175 degrees. In the present invention, each of these signal deflections is handled differently. Although the area of the R wave is large, the area of the noise spike is small because it occurs in a smaller number of samples. Therefore, the characteristic point area is a substitute index for determining which characteristic point is important and which characteristic point is not important.

  The above effect is that small characteristic points having a signal baseline value or a value close thereto are ignored. According to one embodiment, an absolute area threshold in the range of 0.5 is used for the heart rate detection channel.

  In one embodiment, the present invention automatically adjusts the threshold to better adapt to different patients and signals.

  In one embodiment, each characteristic point includes a time component, an amplitude component, and an area component. According to the present invention, time, amplitude, and area components are derived from curvature calculations performed using sample signals.

  A general heartbeat rhythm management device detects a heartbeat by detecting a pulsation in a heartbeat electrical signal and detecting an R wave generation time in the signal. R waves (sometimes referred to as QRS complex) represent the major electrical events for each heart beat. In one embodiment, the heart rate is determined by analyzing a plurality of characteristic points without resorting to beat detection. In one embodiment, an initial estimate of heart rate is obtained by autocorrelating a certain function with characteristic points. In one embodiment, the heart rate is obtained by a time difference autocorrelation function. In one embodiment, a second estimate of heart rate (or number of beats) is obtained by a secondary process and the two estimates are collated and a single value for heart rate is derived.

  In one embodiment, a series of characteristic points of the atrium is generated for the atrial channel and a series of characteristic points of the ventricle is generated for the ventricular channel. The relative timing of a series of characteristic points of the atrium and a series of characteristic points of the ventricle is compared, and the result of this comparison sets the basis for discrimination between different rhythm types. Other methods are also being considered, including identification based on cross-correlation between characteristic point timing and characteristic point regions.

  In one embodiment, a window is set before each ventricular characteristic point. The relative timing is determined by counting the atrial and ventricular characteristic points with respect to the window and graphically displaying the results on a grid. The grid is divided into regions corresponding to different rhythm types.

  For example, the present invention provides the following.
(Item 1)
  An input circuit for receiving a sample signal corresponding to cardiac electrical activity;
  A basic frequency of the sample signal is obtained by autocorrelating a function of a series of characteristic points each having a lobe time in a continuous curve obtained based on the sample signal, A combined controller;
  A memory coupled to the controller and adapted to store the basic frequency;
Implantable heart rate rhythm management device.
(Item 2)
  The apparatus according to item 1, wherein the controller is adapted to obtain a size of the characteristic point and obtain the basic frequency as a function of the size.
(Item 3)
  Further included is a heart rate estimator coupled to the controller and adapted to generate a beat frequency, wherein the controller generates a heart rate based on the beat frequency and the basic frequency. The apparatus according to item 1, characterized in that:
(Item 4)
  The apparatus of claim 1, further comprising a telemetry circuit coupled to the controller and adapted to communicate with a programmer.
(Item 5)
  The apparatus of claim 1, further comprising a treatment circuit coupled to the controller and adapted to apply electrical stimulation in response to a signal received from the controller.
(Item 6)
  Setting a series of characteristic points from the continuous curves generated as a function of the sample input signal, each corresponding to the lobe generation time in the continuous curves;
  Using a processor to determine the frequency associated with the input signal by autocorrelating the function of the series of characteristic points;
  Storing the frequency in a memory;
Including methods.
(Item 7)
  The method according to claim 6, wherein the generation time of the lobe includes the generation time of the center of the lobe.
(Item 8)
  For each characteristic point, the step of determining a size as a function of the area of the lobe is included, and further using the processor to determine a frequency for the input signal, as a function of the size of each characteristic point, Item 7. The method according to Item 6, comprising the step of determining the frequency.
(Item 9)
  7. The method of claim 6, wherein autocorrelating the series of characteristic point functions includes autocorrelating a series of time differences for adjacent characteristic points as a function of occurrence time for each characteristic point. .
(Item 10)
  As a function of the time of occurrence for each characteristic point, the step of autocorrelating a series of time differences for adjacent characteristic points includes obtaining a product value of the time difference between adjacent characteristic points and the time difference between adjacent characteristic points shifted in time. 9. A method according to item 8, characterized in that it is included.
(Item 11)
  11. The method according to item 10, wherein the adjacent characteristic point has a temporally overlapping relationship with the time-shifted adjacent characteristic point.
(Item 12)
  The method of claim 6, wherein autocorrelating the function of the series of characteristic points includes autocorrelating a product of at least two elements selected from the series of characteristic points.
(Item 13)
  Receiving a sample input signal as a function of sensed cardiac electrical activity;
  Utilizing a processor to generate a continuous curve as a function of each received sample;
  As a function of the continuous curvature, each is associated with a lobe in the continuous curvature, each having a series of times as a function of the lobe occurrence time and a size as a function of the lobe area. Generating a characteristic point;
  Autocorrelating a function based on the series of characteristic points to obtain a basic frequency;
  Storing the basic frequency in a memory coupled to the processor;
Including methods.
(Item 14)
  14. The method of item 13, wherein the step of autocorrelating the function includes autocorrelation in the time domain.
(Item 15)
  14. The method according to item 13, wherein the step of autocorrelating the function includes autocorrelation in a characteristic point region.
(Item 16)
  Receiving the sample signal includes receiving a ventricular electrocardiogram, and further, autocorrelating the function based on the series of characteristic points includes time domain autocorrelation of characteristic point time differences. 14. The method according to item 13, characterized in that:
(Item 17)
  Receiving the sample signal includes receiving a defibrillation channel electrogram;
14. The method of item 13, wherein the step of autocorrelating the function based on the series of characteristic points includes time domain autocorrelation of a characteristic point time difference function.
(Item 18)
  Furthermore, the method of the item 13 characterized by including the step of giving a treatment according to the said basic frequency.
(Item 19)
  Means for receiving a sample input signal from cardiac electrical activity;
  Means for generating a continuous curvature based on the input signal;
  Means for identifying a series of lobes in the continuous curve;
  Means for determining the occurrence time of the area center for each lobe in the continuous curve;
  Means for setting a series of characteristic points, each corresponding to one lobe;
  Means for determining a basic frequency for the input signal by autocorrelating the function of the series of characteristic points;
Including system.
(Item 20)
  Item 20. The system of item 19, wherein the receiving means includes a cardiac electrode.
(Item 21)
  Item 20. The system according to Item 19, wherein the means for generating the continuous curve includes a processor.
(Item 22)
  Item 20. The system of item 19, wherein the means for determining the occurrence time includes means for calculating the area of each lobe.
(Item 23)
  further,
  Secondary means for determining a second frequency as a function of the input signal;
  Collation adjusting means for generating a pulsation frequency as a function of the basic frequency and the second frequency;
The system according to item 19, characterized in that it includes:
(Item 24)
  By autocorrelating the function of the series of characteristic points, the means for determining the basic frequency for the input signal by autocorrelating a series of time differences for adjacent characteristic points as a function of the time of occurrence for each characteristic point. The system according to item 19, further comprising means for determining a basic frequency related to the input signal.
(Item 25)
  With relevant data,
  When the data is accessed, the result is to the machine:
  Receiving a sample signal; and
  Generating a continuous curvature based on the sample signal;
  Generating a series of characteristic points in the sample signal, each corresponding to one lobe in the continuous curve and having a time corresponding to the time of occurrence of the lobe;
  Determining the frequency by autocorrelating the function of the series of characteristic points;
  Storing the frequency in a memory;
A product containing a machine-accessible medium that will cause the
(Item 26)
  Item 26. The product of item 25, further comprising, when the data is accessed, causing the machine to generate a size determined as the size of the lobe area in the continuous curve for each characteristic point.
(Item 27)
  Item 26. The step of autocorrelating the function of the series of characteristic points includes the step of autocorrelating the function of the series of characteristic points with a size exceeding a predetermined value. Product.
(Item 28)
  A ventricular channel input circuit for receiving a sample ventricular input signal over a period of time;
  An atrial channel input circuit for receiving a sample atrial input signal over said period;
  A controller coupled to the ventricular channel input circuit and coupled to the atrial channel input circuit and adapted to determine a relative timing between the ventricular characteristic point and the atrial characteristic point, each characteristic point being A controller determined based on a continuous curve generated as a function of the input signal and having a time of one lobe in the continuous curve;
  An identification circuit coupled to the controller and configured to classify the time period based on a comparison of a first function of the atrial characteristic point and a second function of the ventricular characteristic point;
Implantable heart rate rhythm management device.
(Item 29)
  29. The apparatus of item 28, wherein the identification circuit is adapted to generate a diagram of a first function of the atrial characteristic point and a second function of the ventricular characteristic point.
(Item 30)
  Item 30. The apparatus according to Item 29, wherein the identification circuit sets a separation curve in the figure.
(Item 31)
  Item 30. The apparatus according to Item 29, wherein the identification circuit sets a separation line in the figure.
(Item 32)
  Item 32. The apparatus according to Item 31, wherein the identification circuit sets a separation region close to the separation line.
(Item 33)
  29. Apparatus according to item 28, wherein the controller is adapted to determine relative timing by generating a window before each ventricular characteristic point.
(Item 34)
  Item 30. The apparatus of item 30, wherein the size of the window is selectable at a remote location.
(Item 35)
  The apparatus according to item 30, wherein the size of the window is changed according to the measured parameter.
(Item 36)
  The identification circuit obtains a subtraction count based on a ventricular characteristic score having at least one atrial characteristic point in the window, and an atrial characteristic score located in at least one window of the ventricular characteristic point 29. The apparatus according to item 28, wherein an addition count is obtained based on
(Item 37)
  Each characteristic point includes a size determined as a function of the area of the lobe, and the identification circuit classifies the time period as a function of the size of each characteristic point. The device described in 1.
(Item 38)
  29. The apparatus of item 28, further comprising a memory coupled to the controller and adapted to store the classification result for the period.
(Item 39)
  29. The apparatus of item 28, further comprising a telemetry circuit coupled to the controller and adapted to communicate with a programmer.
(Item 40)
  29. The apparatus of item 28, further comprising a treatment circuit coupled to the controller and adapted to apply electrical stimulation in response to a signal received from the controller.
(Item 41)
  A first series of first characteristic points are set from a first continuous curve generated as a function of the first sample input signal of the heartbeat signal and a function of the second sample input signal of the heartbeat signal Setting a second series of second characteristic points from a second continuous curve generated as such that each characteristic point corresponds to a lobe occurrence time in the continuous curve;
  Using a processor to determine the temporal proximity of the second characteristic point to the first characteristic point;
  Classifying the signal according to the temporal proximity;
  Storing the classification result; and
Including methods.
(Item 42)
  42. The method of item 41, wherein the lobe occurrence time includes an occurrence time of the lobe center.
(Item 43)
  42. The method of item 41, wherein the first input signal is received from a first ventricle and the second input signal is received from a second ventricle.
(Item 44)
  42. The method of item 41, wherein the first input signal includes a ventricular input signal and the second input signal includes an atrial input signal.
(Item 45)
  42. A method according to item 41, wherein the classifying step includes the step of setting a plurality of windows, each positioned before each first characteristic point.
(Item 46)
  In the step of classifying the heartbeat signal,
  Determining a subtraction count as a function of the number of first characteristic points having at least one second characteristic point within one of the plurality of windows;
  Determining an addition count as a function of the number of second characteristic points located within at least one of the plurality of windows;
  Drawing the subtraction count as a function of the addition count;
46. A method according to item 45, comprising:
(Item 47)
  47. The method of item 46, wherein plotting the subtraction count as a function of the addition count includes plotting the subtraction count divided by the total number of the first characteristic points.
(Item 48)
  In the step of plotting the subtraction count as a function of the addition count, the addition count divided by the ratio of the number of the first series of first characteristic points and the number of the second series of second characteristic points. 47. A method according to item 46, comprising the step of drawing.
(Item 49)
  47. A method according to item 46, wherein the step of plotting the subtraction count as a function of the summation count includes plotting the summation count expressed as a percentage.
(Item 50)
  Means for receiving a first sample input signal and a second sample input signal, each of which is a function of cardiac electrical activity over a period of time;
  Means for generating a first continuous curve based on the first sample input signal and a second continuous curve based on the second sample input signal;
  Means for identifying a series of lobes in each continuous curve;
  Means for determining the lobe generation time in each continuous curve;
  Each characteristic point corresponds to one lobe, and means for setting a first series of characteristic points based on the first continuous curve and a second series of characteristic points based on the second continuous curve When,
  Means for correlating the time of occurrence of the first series of characteristic points and the time of occurrence of the second series of characteristic points;
  Means for classifying the cardiac electrical activity based on the correlation;
Including system.
(Item 51)
  51. The system of claim 50, wherein the means for determining the lobe occurrence time includes means for determining the lobe center occurrence time.
(Item 52)
  51. The system of item 50, wherein the means for determining the occurrence time includes means for determining a center size of the lobe, wherein the size is a function of the lobe area.
(Item 53)
  51. A system according to item 50, wherein the correlating means includes means for setting a plurality of time windows, each arranged before the first series of characteristic points.
(Item 54)
  Means for obtaining a subtraction count based on the number of first series of characteristic points having at least one of the second series of characteristic points within one of the plurality of time windows in the correlation means; Item 53. The item 53 further comprising means for determining an addition count based on the number of second series of characteristic points disposed within at least one window of the first series of characteristic points. system.
(Item 55)
  55. A system according to item 54, wherein the classifying means includes means for plotting a function of the subtraction count and the addition count.
(Item 56)
  56. A system according to item 55, wherein the classifying means includes means for comparing distributions with respect to a separation curve in a diagram of subtraction count and addition count functions.
(Item 57)
  51. A system according to item 50, wherein the receiving means includes a cardiac electrode.
(Item 58)
  With relevant data,
  When the data is accessed, the result is to the machine:
  Receiving a first sample signal and a second sample signal, each based on cardiac electrical activity over a period of time;
  Generating a first continuous curve and a second continuous curve based on the sample signal;
  A first series of characteristic points in the first sample signal, each corresponding to one lobe in the continuous curve and having a time corresponding to the time of occurrence of the lobe, in the second sample signal Generating a second series of characteristic points;
  Generating a time period classification based on a diagram of the temporal occurrence of the first series of characteristic points compared to the temporal occurrence of the second series of characteristic points; and a separation curve;
  Storing the classification result in a memory;
Products that contain machine-accessible media.
(Item 59)
  Item 58. When the data is accessed, further results in causing the machine to generate a size determined as a function of the lobe area in the continuous curve for each characteristic point. Product.
(Item 60)
  When the data is accessed, the result is further causing the machine to generate a plurality of windows, each positioned before one characteristic point of the first series of characteristic points. Item 65. The product according to Item 58,
(Item 61)
  Plotting a subtraction count based on the number of first series of characteristic points comprising at least one of the second series of characteristic points in one of the plurality of windows in the class generation step; And a means for determining an addition count based on the number of second series of characteristic points arranged in at least one window of the first series of characteristic points. 60. The product according to 60.

FIG. 6 illustrates an arbitrary signal as a function of time. FIG. 6 illustrates a contact circle adapted to an arbitrary signal as a function of time. It is the figure which illustrated the signal expressed with the value of the analog-digital converter as a function of the number of samples. FIG. 6 is a diagram illustrating a signal having a certain curvature expressed as an analog-to-digital converter value as a function of the number of samples. FIG. 6 illustrates a curve signal lobe as a function of time. It is the figure which illustrated the method for calculating | requiring the area under a curved lobe in the graph. FIG. 6 is a characteristic point region autocorrelation diagram of characteristic points according to one embodiment of the present invention. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. FIG. 5 is a diagram illustrating a method for generating a characteristic point time difference function according to one embodiment of the present invention and the resulting time domain autocorrelation. It is a characteristic point area | region autocorrelation figure derived from a series of characteristic points. It is the figure which illustrated the fraction peak of the characteristic point area | region by subharmonic. It is the figure which illustrated the time function showing the space | interval between the characteristic points before and behind time shift. It is the figure which illustrated the XY plane corresponding to the time relationship between the atrial characteristic point and the ventricular characteristic point regarding the selected rhythm type. FIG. 6 illustrates a method for performing identification between rhythms according to one embodiment of the present invention. 1 is a block diagram of an implantable device according to one embodiment of the invention. FIG.

  Other aspects of the invention will become apparent by reading the following detailed description of the invention and examining the drawings that form a part thereof.

  In the drawings, like numerals represent substantially similar components throughout the several views. Similar numbers with different subscripts represent different examples of substantially similar components.

  In the following detailed description, reference is made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration specific embodiments in which the invention may be practiced. These embodiments have been described in sufficient detail to enable those skilled in the art to practice the invention, but of course, these embodiments can be combined or depart from the scope of the invention. Alternatively, other embodiments may be utilized and structural, logical, and electrical changes may be made. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope of the present invention is defined by the appended claims and their equivalents.

Disclosed herein are systems, methods, and apparatus that require a heart rate rhythm management system utilized to detect heart rate signals, including intracardiac electrograms or surface electrocardiograms (ECGs), among others. Such heart rate rhythm management systems include, but are not limited to, pacemakers, CRT devices, cardioverter / defibrillators, pacers / defibrillators, drug delivery devices. Needless to say, the signal compression method and apparatus are not limited to heartbeat detection and other related to cardiac activity, including but not limited to mechanical motion, sound, pressure, acceleration, and impedance signals. It is possible to apply to the signal.
Analysis Based on Curvature For any voltage versus time signal, it is possible to capture significant characteristics by the point that indicates the characteristic “turn” along the signal. In general, these points are different from those selected using the first or second derivative criteria shown in FIG. In this diagram illustrating an arbitrary electrogram 100, the R wave represented as point 101 has a high second derivative d ² V / dt ² value. Furthermore, the electrogram sections on either side of the point 101 have positive and negative steep slopes that can be detected on the first derivative basis. However, these steep slopes (or even the steepest slopes) do not depict a particularly prominent part of the electrogram. For example, for each time point along the segment between points 101 and 102, the waveform has a negative steepness, but none of the points along this segment is more noticeable than the other points.

  On the other hand, the points 102, 103, 104, and 105 are not the maximum value, minimum value, or derivative thereof of the electrocardiogram. These points represent the shape of an arbitrary waveform. Points 102, 103, 104, and 105 are prominent because they indicate positions where the signal turns greatly. The turn at point 101 is very abrupt, and the turns at points 103 and 105 are less abrupt and wider. The turns at points 102 and 104 are more gradual but fairly local. In the present invention, the points 101, 102, 103, 104, and 105 are detected using a criterion based on the curvature of the signal.

  Based on one embodiment of the present invention, signal curvature is illustrated in FIG. In this figure, an optional electrogram 120 is shown. For each point along the electrogram 120, it is possible to find a contact circle that is tangential to the local part of the electrogram. Since the curvature of the electrogram 120 at that point is the reciprocal of the radius, the curvature of the small circle is large, but the curvature of the large circle is small. FIG. 2 shows these circles at selected points. The circle at point 122 is larger than points 121, 123, and 124, and therefore the curvature is small. The turns at points 121, 123, and 124 are more abrupt and the curvature increases.

  In general, the curvature at a point (X, Y) of an arbitrary curve in a two-dimensional space is expressed as follows.

  As you can see, curvature is a non-linear combination of both the first and second derivatives of the curve. At the point where the first derivative along the curve is zero (eg, point 101 in FIG. 1), the curvature is equal to the second derivative and the second derivative is zero (eg, regardless of its slope, In any straight section), the curvature is zero.

  In the present invention, the curvature of an arbitrary input signal is calculated for each sample.

  Let us consider the problem of the dimensionality of curvature with respect to the electrogram. If both X and Y have the same dimension (eg, length), the curvature dimension is 1 / length. In the case of a signal V (t) having a voltage as a function of time as in the electrogram, the signal is time-to-time signal T (t) converted according to T (t) = V (t) / U. Here, U is a constant having a dimension of voltage / time. With this transformation, the first and second derivatives of T (t) are as follows:

This is dimensionless,

This dimension is 1 / time, so the curvature dimension is 1 / time. Accordingly, the curvature is expressed as follows.

This dimension is 1 / hour and U has a numerical value.

  Next, consider curvature as a function of signal gain or amplitude of the input signal. It is assumed that an arbitrary gain G is applied to the input signal in order to obtain a new input signal F (t). Here, F (t) = G · T (t) = V (t) · G / U. The curvature of this gain signal is therefore:

  The G / U ratio can be expressed as W having a time / voltage dimension. In one embodiment, the input signal is a voltage sampled by a 12-bit analog-to-digital converter (ADC) with voltage values in the range of +/− 2048, where each value is a number of This represents the basic amplitude unit ΔV, that is, the voltage resolution. In one embodiment, the amplifier of the present invention is adjusted so that the samples from V (t) approximately meet this range. Furthermore, it is assumed that the samples are extracted at a constant rate, so that time is represented by an integer number of samples, each sample corresponding to a time interval ΔT = time resolution = 1 / (sampling rate).

  FIG. 3 illustrates an XY numeric space 130 represented by the number of samples on the X axis and the analog / digital conversion value on the Y axis. Consider a square 135 in the XY space 130 of 256 steps along the X axis and 256 steps along the Y axis, and a rectangle 140 in the voltage / time space 145 represented by this square. The rectangle 140 has a width of 256 / (sampling rate) seconds and a height of 256 / (voltage resolution) volts. Depending on the values selected for ΔV and ΔT, this rectangle may or may not be square in voltage / time space.

  In one embodiment, W is selected that requires a square in voltage / time space to be represented by a square in sample / sample space. Under that condition, the relationship between curvature and time existing in the voltage / time space is preserved in the sample / sample space.

  FIG. 4 shows an artificial voltage / time curve that forms a semicircle within a square box 150. The curvature of the signal across the box 150 from left to right in voltage-time space is zero until a semicircle is encountered. The curvature increases rapidly to a constant value equal to the radius of 1 / circle, and then suddenly decreases to 0 again at the end of the semicircle. For a particular W value, the representation of this signal in the sample / sample space 155 still has a constant curvature, but for other W values, the curvature will not be constant in the sample / sample space. Let's go.

To maintain this relationship between voltage / time space and sample / sample space, W is chosen as follows. In voltage / time space, a box of width T × height V is considered to be a square. In the sample-sample space, the box has a time sample width of T / ΔT and a voltage sample height of GV / ΔV, and the conversion converts the voltage sample into a time sample using U, and as a result, The square dimension of the sample space is T / ΔT × VG / ΔV / U. If the box is square in the sample space, assuming that the box in voltage / time space (V × T) is square, then: (V / ΔV) · G / U = T / ΔT or G / U = W = (T / ΔT) / (V / ΔV)
According to one embodiment of the present invention, the curvature is based on the first and second derivatives of the signal. By utilizing least squares cubic polynomial fit, the noise that would otherwise result from finding the curvature using a derivative estimate and using non-linear calculations is reduced.

  Since the sample voltage signal is expressed as V (t) = V (I · ΔT) (where t = j · ΔT), 2N + 1 voltage samples centered at time t are used for size N adaptation. Is used, and therefore:

V ([in] ・ ΔT), ... V ([i-2] ・ ΔT), V ([i-1] ・ ΔT), V (i ・ ΔT), V ([i + 1] ・ ΔT ), V ([i + 2] ・ ΔT), ... V ([i + n] ・ ΔT)
Using the set of N sample data points, the least squares cubic fit shown below is performed.

Here, Ai, Bi, Ci, and Di are coefficients obtained by minimizing the square error related to the fit, and dt represents a time step away from i · ΔT from which the value of V _est is obtained. ing. The coefficients of the polynomial are displayed with i to indicate that they are valid for the point i · ΔT.

  Using the above equation, the curvature between two points at the time point i · ΔT is as follows.

  However, the sample point is not necessarily located at the time when the value of the signal curvature becomes maximum or minimum. Thus, in one embodiment, the curvature signal is integrated between adjacent sample points using estimates of the first and second derivatives of the signal at the sample points.

  Furthermore, the following average point curvature formula is obtained by simplification.

  As described above, the curvature is calculated for each sample from the input signal.

  Next, a procedure for obtaining characteristic points in a signal will be considered.

  The original signal turn is reflected as a deviation greater than zero and a deviation less than zero in the curvature signal. As suggested by curve 160 in FIG. 5, lobes above zero (eg, lobe 165) or lobes below zero (eg, lobe 170) thus represent a single turn in the input signal. The opposite curve lobe indicates the opposite turn (left or right) in the signal. The area under each lobe reflects all angles included in the turn.

  Using the point-to-point method, when a lobe is generated, the lobe is identified, and the area and center of each lobe are obtained.

A curvature value is generated for each sample. For each sample time, CRV represents the current curvature value and CRV _old represents the previous value held from the previous sample. When the curvature value is 0, the original signal does not turn, so there is no possibility that a characteristic point exists. The CRV value is almost never equal to zero.

In one embodiment, a dead zone around 0 is defined that can be processed with the calculated curvature equal to 0. While the curvature is within this dead band, the signal hardly turns. The dead zone is defined by the curvature threshold and ranges from + CRV _thresh to -CRV _thresh above and below zero.

In the present invention, in one embodiment, nine cases are identified considering the CRV value relative to the threshold, the absence or direction of lobes. These cases can be defined as follows:
Case 1: CRV> CRV _thresh and not lobed.

In this case, the current curvature value exceeds the dead zone, and the bending signal is not currently in a lobe. Thus, the positive lobe has just begun, so the positive lobe initialization calculation will follow.
Case 2: CRV _thresh ≧ CRV ≧ −CRV _thresh , and no lobes are formed.

In this case, the current curvature value is in the dead zone and the curving signal is not currently in a lobe.
Case 3: CRV <−CRV _thresh and not lobed.

In this case, the current curvature value is less than the dead zone and the curving signal is not currently in a lobe. Thus, the negative lobe has just begun, so the negative lobe initialization calculation will follow.
Case 4: CRV> CRV _thresh and no positive lobe.

In this case, the current curvature value exceeds the dead zone, and the curvature signal does not form a positive lobe. Therefore, the positive lobe continuation calculation follows.
Case 5: CRV _thresh ≧ CRV ≧ −CRV _thresh and a positive lobe is formed.

In this case, the current curvature value is in the dead zone and the curvature signal is currently in a positive lobe.
Therefore, the positive lobe has just ended, and therefore the positive lobe termination calculation will follow.
Case 6: CRV <−CRV _thresh and a positive lobe.

In this case, the current curvature value is less than the dead zone and the curvature signal is currently in a positive lobe. Accordingly, since the positive lobe has just ended and the negative lobe has just started, the positive lobe end calculation and the negative lobe initialization calculation will follow.
Case 7: CRV> CRV _thresh and a negative lobe.

In this case, the current curvature value exceeds the dead zone, and the bending signal has a negative lobe. Therefore, since the negative lobe has just ended and the positive lobe has just started, the negative lobe termination calculation and the positive lobe initialization calculation will follow.
Case 8: CRV _thresh ≧ CRV ≧ −CRV _thresh and a negative lobe is formed.

In this case, the current curvature value is in the dead zone and the curvature signal has a negative lobe. Thus, the negative lobe has just ended, so the negative lobe termination calculation will follow.
Case 9: CRV <−CRV _thresh and a negative lobe.

  In this case, the current curvature value is less than the dead zone and the curvature signal has a negative lobe. Therefore, the negative lobe continuation calculation follows.

  In one embodiment, hysteresis is utilized to help identify curved lobes. In one embodiment, when the lobe starts, the curvature value is set to intersect a threshold closer to 0 to end the lobe. Thus, another threshold is introduced by hysteresis.

  Next, consider the metrics for characterizing each lobe according to the present invention. In one embodiment, these metrics include the total lobe area, the lobe area center time, and the original data value at the area center time. In one embodiment, other metrics are utilized, including, for example, peak curvature in the lobe, time of peak curvature, lobe start and lobe end time.

FIG. 6 shows an example of a curved lobe 185 as a series of curvatures calculated at each sample time. Tick marks 180 along the X axis represent actual samples and are therefore separated by ΔT. Zero curvature is shown as the X axis. Curvature thresholds + CRV _thresh 190 and -CRV _thresh 195 are shown as horizontal lines above and below the X axis. Inflection points 200-245 are shown as small solid circles.

As shown in this figure, the initial curvature values at the inflection points 200 and 205 are within the dead zone of the curvature threshold, and therefore no lobes have yet been formed. The bending point 210 exceeds the threshold value, and corresponds to the above-described case 1. In accordance with Case 1 above, the magnitude of the initial area 250 is calculated as soon as the lobe starts. Once the inflection point 215 is determined, the contribution to the size of the main area 255 is calculated according to Case 4 above. Furthermore, when the bending points 220, 225, 230 are determined, the size of the main area 255 is increased. Once the bending point 235 below + CRV _thresh 190 is determined, the size of the final area 260 is calculated for the area below the bending lobe 185 and between the bending point 230 and the intersection of + CRV _thresh 190. . Inflection points 235, 240, and 245 are in the dead zone and do not contribute to area calculation. The area of the curved lobe 185 including the sum of the initial area 250, the main area 255, and the final area 260 is calculated.

  In one embodiment, the counter value M is maintained by the processor to monitor the progress of each lobe sample. When the “start lobe” calculation is performed, the value of M is reset to zero and incremented for each subsequent point that remains outside the deadband. The M value is illustrated at different inflection points in FIG.

To determine the area under the curved lobe, the M value is set to 0, and when the lobe begins, the initial area of the curved lobe is calculated as follows: Initial area = (C _i + CRV _thresh ) · (C _i −CRV _thresh ) / (C _i −C _i−1 ). For each sequential curvature value, the M value is incremented and the additional area contribution is calculated as follows: Additional area =-(M-1) · C _i + M · C _i-1 . When the lobe ends, the M value is not incremented and the final area contribution is calculated as follows: Final area = M · C _i-1 + (C _i + CRV _thresh ) · (C _i -CRV _thresh ) / (C _i -C _i-1 ). As a result, the area under the curved lobe can be obtained by multiplying the sum of the initial, final, and total additional areas by ΔT / 2.

A similar approach is used to determine the first moment of area under the curved lobe. When calculating the initial area of the curved lobe, the initial moment of the area is also calculated as follows. Initial area = − {C _i + 2CRV _thresh } · γ ² . Where γ = (CRV _thresh -C _i-1 ) / (C _i -C _i-1 ). For each sequential curvature value in the lobe, the additional moment contribution is calculated using the same M value as the area calculation. This additional moment contribution is calculated as follows. Additional moment =-{C _{i ·} (3M-1) + C _{i-1 ·} (3M-2)}.

When the lobe ends, the final moment contribution is calculated simultaneously with the final area calculation. The final moment is calculated as follows: Final moment = [CRV _thresh · {3M + 2γ} + C _i-1 · {3M + γ}] · γ.

As a result, initial, final, the sum of all additional moment is multiplied by [Delta] T ^2/6, it is possible to determine the first moment of area of curved lobe lower. The time at the center of the curved lobe area is determined by dividing the primary moment of the curved lobe area by the curved lobe area, where time = primary moment / area. This time relates to the time of the inflection point where the lobe started.

  Calculations for both area and area first moment are done in such a way that once the samples are collected, the contributions of each curvature sample can be summed to a total area or total first moment of area. Therefore, all the calculation elements include the current and previous curvature values, threshold values, and counter M. Thus, as soon as the lobe is finished, the area and the center of the area are generated.

  In one embodiment, in the present invention, characteristic points are identified from processing for each point of the input signal. In one embodiment, signal characteristic points are detected and stored in a buffer. In one embodiment, each characteristic point is a set of values including the time of the characteristic point, the value of the input signal at that time, and the value representing the direction and length of the curve in the signal that produced the characteristic point.

  Next, a method for extracting characteristic points will be considered. In one embodiment, ECG characteristic points are extracted by sampling at 400 Hz. In one embodiment, sampling is performed at 200 Hz. By filtering the extracted data using a 5-point (400 Hz) or 3-point (200 Hz) moving average filter and a 5-least-squares regression average point curvature method, characteristic points related to the QRS complex are selected.

  When detecting curvature associated with a slower form, sampling is performed at a lower rate, for example, 50 Hz. An example of a wave having a slower form would be a T wave in an electrocardiogram.

  In one embodiment, characteristic point extraction is performed using multiple sampling rates.

  In one embodiment, double speed sampling is performed with a sampling rate of 200 Hz resulting in a fast characteristic point and 50 Hz resulting in a slow characteristic point, and a three-point moving average filter is used. Consider an example of dual-rate sampling using 200 Hz that yields a fast characteristic point and 50 Hz that yields a slow characteristic point by 5-point least square regression size and 3-point moving average filtering at both sampling rates.

In one embodiment, the curvature is determined using 5-point least squares regression filtering. The coefficients for obtaining the straight line (Bi) and parabola (Ci) fitting coefficients in the cubic least square regression with respect to the data (D _{i-2 to} D _{i + 2} ) at the point i are as follows.

Here, Rate is a sampling rate of display in samples / second, and Bi is five corresponding data centered on i (D _{i−2 to} D _{i + 2} ) with these least square regression coefficients P. The product sum multiplied by the points, and Ci is the sum using the least square regression coefficient Q. Accordingly, the curvature is calculated as follows.

Here, W is a constant. Since W appears in relation to the fitness coefficient Ci or Bi, its value can be incorporated into the calculation as a data signal gain such that D ′ (i) = W · D (i). Therefore, the curvature equation is as follows.

Here, B'i and C'i are obtained by using D 'data points together with least square regression coefficients instead of D data points.

  Therefore, the curvature equation can be rewritten as follows.

Here, Gn is an arbitrary value incorporated in C ″ (along with 2). If the curvature threshold used to detect the characteristic points is adjusted accordingly, the curvature for each point is calculated. It does not matter whether Gn × curvature of each point is calculated. The value 2 · Gn is incorporated into Ci ″ by multiplying 2Gn by a new least-square regression coefficient Q _{−2 to} Q ₂ . This is implemented by changing to a least square regression coefficient (Q ′ _{−2 to} Q ′ ₂ ). If Gn is selected as 7 / (6Rate) and the term Fn = Rate / 12 is incorporated as an additional data signal gain, the least squares regression coefficient is

As a result, D ″ (i) = 2 · W · D (i), Ci ″ = Q ″ ₋₂ · D ″ (i−2) + Q ″ ₋₁ · D ″ (i−1) + Q ″ ₀ · D "(I) + Q" _{+ 1.D} "(i + 1) + Q" _{+ 2.D} "(i + 2), Bi" = P " _-2.D " (i-2) + P " _-1.D " (i- 1) + P ″ ₀ · D ″ (i) + P ″ ₋₁ · D ″ (i + 1) + P ″ _{+ 2} · D ″ (i + 2)

  Note that these coefficients are all powers of 2 (0, 1, 2, 4 or 8), thus simplifying the hardware or firmware implementation of the present invention.

  When a characteristic point is obtained using two sampling rates, the calculated value of the curvature will be different if calculated at different sampling rates. Therefore, two threshold levels and a set of parallel calculation values are used. In addition, the moving average filtering uses five consecutive points in the case of high-speed filtering, but in the low-speed filtering, there is a difference that every third point is used at intervals of 16 points. is there.

  In one embodiment, the linearity of least squares regression operations is utilized. The moving average filtering operation is performed at the fitness coefficient level (ie, B and C), not at the data level. For this embodiment, two filtering operations (one for B and one for C) are performed at each rate.

  In one embodiment, fast and slow moving average filtering operations are performed in parallel.

  In one embodiment, the value for the characteristic point is selected using the following method.

  Starting from the time of one sample, the detection hardware and software of the present invention begins collecting the next sample so that the actual sample reflects the signal characteristics between the previous sample and the current sample. To do. In the real time of the sample, the sample value can be attributed to an intermediate time between the current sample and the previous sample. A sample with a low sampling rate applies to a signal with high-speed sampling two stages before the actual sample time.

  In one embodiment, a circular data buffer is used. Data is collected at a high sampling rate and used to fill the circular data buffer. When the timing of the high-speed or low-speed characteristic point is determined, the data value related to the characteristic point is obtained from the data stored in the circular buffer. Therefore, the data buffer is given a size such that necessary data can be obtained in the buffer. In one embodiment, the buffer is sized to hold approximately 50 milliseconds of data or 256 data points. In one embodiment, the circular buffer is implemented as a memory array with an index pointer. In one embodiment, the index pointer includes a binary counter and the buffer size is a power of two.

  In one embodiment, moving average filtering (RAF) is performed on the input data. In various embodiments, the filtering is performed by a circular buffer or a shift register set. The value of the circular data buffer reflects the fast moving average filtering. In one embodiment, a hybrid approach is used that performs fast moving average filtering and then stores the data in a circular buffer.

  In one embodiment, moving average filtering reduces curvature noise associated with the signal content of the fast curvature signal. An electrocardiogram signal generally deviates greatly from zero curvature during the fast turn of the signal. However, if the fast turn is not so severe, the curvature signal will not be too far from zero, and curvature noise can cause the curvature lobe to end prematurely. As a result, one small curved lobe is divided into two or more smaller lobes. Thus, in one embodiment, curvature noise associated with signal content is reduced by moving average filtering for fast curvature signals.

  In the filtering of the input signal, the same effect as the filtering of the actual bending signal cannot be obtained. In one embodiment, three-point moving average filtering is applied to the fast curvature values.

  In one embodiment, the fast curvature value is calculated as the average point curvature and three-point moving average filtering, so the fast curvature calculated at the time of taking the current sample is the input signal 5 hours before the current sample. Applies to time.

  In one embodiment, the low-speed curvature value is calculated at 1/4 of the high-speed sampling rate and calculated as the average point curvature, so if every third high-speed sample is taken, the low-speed curvature is calculated and the current sample This applies to the time of the input signal prior to a lower 4 (16 higher) time step.

  In one embodiment, the detection of curved lobes and characteristic points requires both high-speed characteristic point detection and low-speed characteristic point detection performed in parallel. In one embodiment, separate curvature and curvature area thresholds are utilized for high speed and low speed operation.

  In one embodiment, the bending lobe ends when the bending signal is in the opposite direction to the crossing direction that started the lobe, but crosses the same threshold used to start the lobe. In one embodiment, the bending lobe ends when the bending signal crosses zero as the hysteresis value.

  In one embodiment, the W value is selected according to the following procedure. The recent range of W is selected by the following analysis. At a sampling rate of 20 Hz, QRS takes approximately 5 samples. The peak signal is approximated by 128. If QRS reaches a peak from 0 at 5 points and is further approximated as a semicircle with an arch shape returning to 0 (in curved space), the radius of the semicircle will be 0.01 seconds and the peak will be The value is 0.01. Therefore, when the W value is multiplied, the value of the peak (128) becomes 0.01, that is, W = 0.01 / 128 = 0.0078. In one embodiment, W = 0.000012 = 8192, which is a power of 1/2.

  Consider the selection of curvature for characteristic point detection. In one embodiment, the threshold value reduces or eliminates noise characteristic points and computational burden. This noise is a natural consequence of the curvature signal wandering around 0 (or some other value) when the raw signal does not characteristically turn.

  In one embodiment, with respect to the detected curved lobe, the point-by-point curvature will be outside the dead zone formed by thresholds on both sides of zero. If this threshold value is reduced toward 0, the number of detected curved lobes increases and the computational burden increases. In one embodiment, the threshold is selected by creating a histogram of fast curvature values calculated at each sample time using an input signal of about 1 minute in time. The threshold is selected as a value corresponding to 20% of the peak. This value is then used as a threshold for both the fast curve and slow curve feature selection processes.

  In one embodiment, the noise lobe is eliminated by requiring that the detection lobe must exceed a critical area limit. The area of the noise lobe is generally small, but the actual lobe associated with the characteristic turn of the signal is usually quite large.

  In one embodiment, different values are used for the fast curve and slow curve area thresholds. Using the number of beats that occur during the one minute time period used to set the gain and curvature thresholds, the desired number of high and low characteristic points to be determined for these beats is selected. For example, in one embodiment, five high speed characteristic points per beat and eight low speed characteristic points per beat are used as targets. The area threshold is then adjusted until these target number characteristic points are found. In one embodiment, an area lower limit of 0.1 is used.

In one embodiment, a template for classifying beats is created. For example, the characteristic point template includes a rectangular box that contains time (relative to the reference characteristic point) and signal amplitude. The template also includes a curved area code.
Heart rate estimation The heart rate expressed in units of the number of beats per minute can be obtained without depending on the detection of individual beats. The heart rate is mathematically determined by dividing the characteristic points per minute by the characteristic points per beat. Calculation of characteristic points per minute is simple. The value of the autocorrelation function is obtained, and the characteristic score per beat is obtained.

  In one embodiment, the characteristic score per beat is calculated by performing an autocorrelation function in the time domain. In one embodiment, the characteristic score per beat is calculated by performing an autocorrelation function in the characteristic point domain.

In a series or series of characteristic points representing the input signal, the individual characteristic points are CP ₁ , CP ₂ , CP ₃ ,. . . , It shall be displayed as CP _n. Each characteristic point has a time value, an area (turn area) value, and a voltage value. For the i-th characteristic point, these are respectively displayed as CP _iT , CP _iA , and CP _iV .

  Values for autocorrelation range from +1 to -1. The shift index (k) for autocorrelation represents a shift in the characteristic point region. For example, when k = 0, there is no shift, and when k = 1, the characteristic point is shifted by 1.

In one embodiment, the autocorrelation of the characteristic points in the characteristic point region is multiplied by the same reference value of the corresponding characteristic point for each value of the shift index k and summed to produce a value. It will be necessary. For example, when k = 3, the series of values includes the sum of the product of CP _1V and CP _4V , the product of CP _2V and CP _5V , the product of CP _3V and CP _6V , and the like.

  The results are displayed graphically in the autocorrelation diagram of FIG. In this figure, the vertical axis represents the autocorrelation coefficient based on the value obtained by dividing the autocovariance function by the dispersion function, and the horizontal axis represents the shift index k. In the graph of this figure, on average, there is a peak at a shift index of 5 indicating that each beat is represented by 5 characteristic points.

  Using typical data collected over a period lasting 6.288 seconds, a total of 80 characteristic points were generated when the sampling rate was 12.722 characteristic points / second. In the case of 5 characteristic points per beat, the heart rate is calculated as 152.67 beats / minute.

In one embodiment, autocorrelation is performed sequentially using time differences between characteristic points. For example, the time difference between adjacent characteristic points can be expressed as CP _iDT = CP _iT −CP _i−1T . Further, the area difference between adjacent characteristic points can be represented as CP _iDA = CP _iA −CP _i−1A , and the voltage difference can be represented as CP _iDV = CP _iV −CP _i−1V .

In various embodiments, the difference functions CP _iDT , CP _iDA , CP _iDV are autocorrelated to determine the characteristic points per beat. In various embodiments, CP _iA , CP _iV , CP _iDT , CP _iDA , CP _iDV singly or in combination (eg, product CP _iV × CP _iDT × CP _iA ) per beat. Characteristic points are obtained.

  As an example, the autocorrelation of the characteristic point voltage value in the characteristic point region can be expressed as follows.

Here, N is the number of characteristic points, CPiV AVG is an average value of N characteristic point values, and k is an index in the characteristic point region.

  In one embodiment, the heart rate is determined from the autocorrelation of characteristic points in the time domain. In one embodiment, the heartbeat signal is reconstructed for autocorrelation from a series of characteristic points. In one embodiment, this reconstruction yields equally spaced samples for autocorrelation. In one embodiment, reconstruction and autocorrelation are performed in a closed form using characteristic point voltages and time from a series of characteristic points.

  In one embodiment, only timing information about characteristic points in the heartbeat signal is used for autocorrelation. In one embodiment, the difference between the sequential characteristic point time at a series of characteristic points and the time of each characteristic point at a series of characteristic points is a time difference function. In one embodiment, the time difference function is configured to have equally spaced samples for autocorrelation. In one embodiment, the autocorrelation of the time difference function is performed in a closed form using only the characteristic point times from a series of characteristic points. In general, the first peak of autocorrelation is equal to the reciprocal of the heart rate.

  FIG. 8A illustrates the heart rate signal 30 for a selected time period with characteristic points 305 shown. This time includes 5 heart beats each having 4 to 6 characteristic points. FIG. 8B illustrates the signal amplitude at each characteristic point as a function of time. 8C and 8D illustrate time differences between sequential characteristic points in a series of characteristic points. When plotting in comparison with the time of the characteristic point, a time difference function that can be further autocorrelated is obtained by these time differences. In the case of FIG. 8C, the time difference 310 between the adjacent characteristic points 305A and 305B is converted into the display amplitude by the horizontal line 315. Therefore, the time difference function between the characteristic points 305A and 305B is equal to this line segment 315. FIG. 8D illustrates the time difference of all characteristic points during this period. FIG. 8E illustrates the time difference after removing the characteristic points. The waveform shown in FIG. 8E is depicted as an autocorrelation diagram in FIG. 8F, showing that the first peak occurs at a time of 0.360 seconds, corresponding to a beat number of 166 bpm.

  In one embodiment, the heart rate is determined from the estimated characteristic points per beat and the characteristic point times at a series of characteristic points. In one embodiment, an average interval representing one beat is determined, which is the reciprocal of the heart rate associated with the rhythm. For example, when the estimated value of the number of characteristic points per beat is 5, the average time interval related to rhythm is obtained as follows using the average time interval between each characteristic point pair separated by five characteristic points.

Here, Kp is an estimated characteristic score per beat.

  Other autocorrelations are also considered. For example, in various embodiments, characteristic point voltage values and area values are autocorrelated.

  The heart rate can be obtained by using the subharmonic function in the autocorrelation function. The subharmonic frequency is a multiple of the basic frequency of the heartbeat. For example, if the basic rhythm is monogenic ventricular tachycardia (MVT), the same basic pulsatile shape is repeated at regular intervals. This represents that multiple peaks occur in the autocorrelation diagram in the characteristic point region. For example, the autocorrelation diagram for the MVT rhythm shows a peak at a shift index of 5 as well as subharmonics at 10, 15, and 20 shift indices.

  In one embodiment, the heart rate is determined using subharmonics. A typical autocorrelation for DT is illustrated in graph 330 of FIG. The autocorrelation in this graph is derived from a specific heartbeat period. In graph 330, it is not clear whether the first peak 320 for the corresponding atrial heartbeat period is at k = 5 or k = 6. The position of the second peak 325 derived from the first subharmonic allows this ambiguity to be resolved.

  In the autocorrelation processing method according to one embodiment, the resolution for obtaining the peak is improved by incorporating subharmonics. As in the case of FIG. 9, the autocorrelation has values at k = 1, k = 2, and the like. First, additional points are added to the autocorrelation by adding points with 1/4 step resolution. For example, in the case of FIG. 9, additional points are added to 2.25, 2.5, 2.75, 3.25, 3.5, 3.75, and the like. Autocorrelation values for these additional points are determined by interpolating between the original points. Next, the sub-harmonic autocorrelation value is obtained by averaging the values in the harmonic. For example, the subharmonic autocorrelation value at k = 2 is obtained by averaging the points at 2, 4, 6, etc. of the autocorrelation. At point k = 2.25, the value of the subharmonic autocorrelation is determined by averaging the points at autocorrelation k = 2.25, 4.5, 6.75, etc. FIG. 10 illustrates an example of subharmonic autocorrelation from FIG. In FIG. 10, the peak becomes clearer in this case at k = 5.5 between the original correlation steps k = 5 and k = 6. In one embodiment, a subharmonic autocorrelation value is obtained using an average of the first three subharmonic values. In one embodiment, the speed of the rhythm is estimated using the peak of the subharmonic autocorrelation value.

  The autocorrelation in the time domain can be used to identify the periodicity of the signal. Periodicity can also be identified by performing a time domain autocorrelation of the voltage signal using the characteristic point voltage value. In one embodiment, the time domain signal is represented by a straight line connecting continuous characteristic points. The line segment representation of the signal is then autocorrelated using the line segment integration. The end points of the line segments are characteristic points, thus simplifying the calculation.

For example, CP ₁ is the oldest characteristic point, CP ₂ is the next oldest characteristic point,. . . . Consider the signal represented by N characteristic points. Thus, a closed-form autocorrelation for a particular time delay Q is calculated by sliding each line segment by Q and integrating (for that line segment) the product of that line segment and the signal portion it overlaps. . In order to obtain an autocorrelation value for the time delay Q, values are obtained for all line segments. Further, the same operation is performed for other time delay values, and a time domain autocorrelation (autocorrelation with respect to time delay) value can be obtained.

  The heart rate can also be obtained by autocorrelating a function of characteristic points in the time domain. For example, in one embodiment, closed form autocorrelation is calculated as shown in FIG.

Consider a series of characteristic point time differences CP _{i, T} when there are N characteristic points. Although the time difference function 415A is set, its value at each characteristic point is equal to the time interval between CP _i and CP _{i + 1} , as shown by a thick black line and a black solid circle in the chart 405. The time function 415A is shifted in time by Q as represented by 415B in the chart 410. Each point along the shift function 415B will therefore be located at a point in the non-shift function 415A. For example, a point 420B (obtained from the shift point 420A) in the chart 410 is projected to a point 420C in the chart 405. Similarly, the non-shift function value is projected onto the time shift function. For example, the point 425A of the chart 405 is projected onto the point 425B of the chart 410. The autocorrelation value of this function in the time domain is determined by determining the area under the product of the curve, which corresponds to adding the area contribution for each part between the vertical lines shown in FIG.

  Next, the shift curve 415B of the chart 410 will be considered. Each line segment between vertical lines is temporally bounded by shift characteristic points or non-shift characteristic points. Further, these bounding endpoints each have a value equal to the shift value or the non-shift value. Each line segment is again aligned with the second set of values (shown in chart 405) as an unshifted value (black solid circle) or a shifted value (white circle). The value of the white circle (chart 405) and the value of the white circle (chart 410) are obtained by shifting the characteristic points in time in the forward direction (chart 405) or the reverse direction (chart 410). In one embodiment, the area calculation value is obtained from the shift line segment and the corresponding non-shift line segment. For example, the line segment between 420B and the next point in the shift function 415B will be exactly on the same line segment after the point 430 in the non-shift function 415A. The values of 420B and 430 are known. Since 415B is a time difference function, the period from 420B to the next point is still the value of 420B. Therefore, the contribution to the autocorrelation integral for this line segment is equal to the square of the value at point 420B multiplied by the value at point 430. Autocorrelation is the sum of all line segments.

  It is possible to estimate the heart rate using the time domain autocorrelation of the characteristic point time difference function. Various methods are used for estimating the heart rate. In some methods, the autocorrelation between the 0.15 second and 0.6 second time delays is fitted to a straight line by least squares regression, and this trend line is removed from this same range of autocorrelation data. . In one method, a de-trended autocorrelation point-by-point scan is performed to determine all local peaks. Each local peak is therefore only considered if it is faster than the first subharmonic of the first such peak. For example, if the first peak in the autocorrelation corresponds to a heart rate of 30 bpm, peaks below 150 bpm are not considered. All peaks that have not yet been considered have a specific delay time and autocorrelation value. In one embodiment, the autocorrelation values are used to weight the heart rate associated with those peaks to determine the average heart rate associated with the rhythm.

In one embodiment, the average time lag is calculated as SUM (corr ^* lag) / using the value of the non-trend autocorrelation function in the time lag range of 0.15 seconds to 0.6 seconds and a value greater than zero. It is obtained as sum (corr). Using this average time delay, the heart rate is obtained. In this method, the product of each detrended autocorrelation value between 0.15 seconds and 0.6 seconds and a value greater than 0 is multiplied by the corresponding time delay and summed to give SUM ( corr ^* lag). Further, the sums of the non-trend autocorrelation values that satisfy these criteria are summed to obtain SUM (corr).

  In one embodiment, the heart rate is determined using an average autocorrelation value associated with the peak.

Other methods are also considered to determine the heart rate.
Rhythm Identification In one embodiment, distinction between different rhythm types is performed using characteristic point analysis.

  The rhythm type is related to the relationship between the atrial heartbeat signal and the ventricular heartbeat signal. Accordingly, a series of atrial characteristic points and a series of ventricular characteristic points are generated using the atrial input signal and the ventricular input signal to distinguish between different types. In various embodiments, rhythms can be identified by estimating characteristic point timing variations and cross-correlations in characteristic regions. In one embodiment, identification is performed using the relative timing of characteristic points in the atrial and ventricular heart rate channels.

  In one embodiment, individual characteristic points in the atrial channel and individual characteristic points in the ventricular channel are compared.

  In one embodiment, a time window is set before each ventricular characteristic point. With an appropriately sized time window, the characteristic points of the ventricular waveform should have one or more atrial characteristic points within the time window. Each ventricular characteristic point is considered individually and the number of atrial characteristic points included in such a time window is determined. In one embodiment, due to constraints, each atrial characteristic point is required to be counted only once. This count is called the addition count and is displayed as A: V, but is equal to the number of atrial characteristic points located within the time window of one or more ventricular characteristic points.

  The size of the time window is selected so that an accurate rhythm identification result is obtained. In one embodiment, the time window is positioned 30 ms to 10 ms before the ventricular characteristic point. In one embodiment, the time window is positioned 250 milliseconds to 150 milliseconds before the ventricular characteristic point. Other sizes are also considered. In one embodiment, the time window size can be selected based on manual input or can be automatically selected based on measured or calculated parameters. In one embodiment, the time window size is selected at a remote location and communicated to the implantable device.

Consider the resulting count in the case of a simple conduction rhythm where each atrial depolarization is somewhat delayed and coincides with a ventricular depolarization. In such a case, since most of the atrial characteristic points are counted, the ratio of the counted atrial characteristic points to the total number of atrial characteristic points (indicated by _Acnt ) is approximately 1.

  In one embodiment, the far field detection of ventricular depolarization in the atrial channel may result in additional characteristic points in the atrial channel, which are generally coincident with the ventricular characteristic points, Or it will be later, so it will not fit within the time window.

  If the heart is pulsating due to atrial fibrillation (A-fib) rather than a one-to-one conduction rhythm, the total count of atrial characteristic points is increased. Also, since not all atrial waveforms are conducted and result in a ventricular waveform, the proportion of atrial characteristic points that fall within the time window of the ventricular characteristic points is reduced. Therefore, the ratio should be small.

  If the heart is beating due to MVT, multigenic ventricular tachycardia (PVT), or ventricular fibrillation (VF), the total count of atrial characteristic points will not increase, but AV synchronization The characteristic score associated with the time window of the ventricular characteristic point is reduced due to the loss of performance. On the other hand, more ventricular characteristic points will be provided and the A: V number will increase. Therefore, the ratio is not likely to be greatly affected.

  Next, the same timing is considered from the viewpoint of ventricular characteristic points. Each ventricular characteristic point is analyzed and the number of ventricular characteristic points having at least one atrial characteristic point within the time window of the ventricular characteristic point is counted. This count is called the subtraction count and is represented by V: A, but is equal to the ventricular characteristic score having one or more atrial characteristic points within the preceding time window.

In the case of a one-to-one conduction rhythm, most of the ventricular characteristic points satisfy the standard, so the ratio of the counted ventricular characteristic points to the total number of ventricular characteristic points (V _cnt ) is approximately 1. .

  If the heart rate rhythm is not a one-to-one conduction rhythm but is actually an A-fib, the total number of ventricular characteristic points will reflect the higher heart rate more greatly. However, since each ventricular waveform is due to conduction rhythm, even if the ventricular characteristic points increase, the proportion of ventricular characteristic points related to the atrial characteristic point remains substantially the same.

  If the actual rhythm is MVT or PVT / VF, the total number of ventricular characteristic points will increase and the number associated with atrial characteristic points within those time windows will decrease. Therefore, the ratio also decreases.

Over a selected period, the ventricular and atrial characteristic points are determined. If the rhythm is one-to-one conduction, the same number of beats occur in each channel. Without knowing the number of beats in that period, it is possible to estimate the atrial characteristic score per beat (denoted herein as A _cnt / V _cnt ) as compared to the ventricular characteristic score per beat. .

  If the rhythm is one to one, the proportion of atrial characteristic points associated with ventricular characteristic points should approximate the proportion of ventricular characteristic points associated with atrial characteristic points, but the number of characteristic points per beat is two There may be differences in the channels.

In various embodiments, the value V: A / V _cnt , or a value that can be rewritten as (V _cnt / A _cnt ) ² / (A: V ^* V / A) [(A: V / A _cnt ) / (V: A / V _cnt )] / [A _cnt / V _cnt ] is used to discriminate between rhythms.

In one embodiment, a diagram is generated that graphically illustrates the relationship between the atrial characteristic point and the ventricular characteristic point. FIG. 12 shows a diagram for about 50 rhythms with supraventricular origin (SVT) or ventricular origin (VT). The X axis or horizontal axis is 10 ^* (V: A / V _cnt ), and the Y axis or vertical axis is 10 _* (A: V _* V _cnt / A _cnt ² ).

Different regions in the XY plane correspond to different rhythm types. For example, the SVT rhythm is generally located in the range of less than about 150 on the Y axis and 50 to 10 on the X axis. In contrast, the VT rhythm is generally much closer to zero on the X axis, reflecting the expected significant decrease in V: A / V _cnt . Furthermore, in general, since V _cnt is higher, the VT rhythm has a higher value on the Y axis.

  In one embodiment, the rhythmic X and Y values are determined by processing the timing of the atrial and ventricular characteristic points according to the present invention. In one embodiment, the position of the X and Y values on the graph is used for rhythm classification. In one embodiment, the classification will be SVT or VT. In one embodiment, this is done by comparing the XY point with a fixed boundary on the graph. In one embodiment, the boundary is composed of one or more line segments. In one embodiment, the rhythm is classified as SVT when the XY point is located on one side of the boundary and is classified as VT when located on the other side of the boundary.

FIG. 13 illustrates an exemplary method 450 according to one embodiment of the present invention. At 455, a series of atrial characteristic points are generated from the atrial channel and a series of ventricular characteristic points are generated from the ventricular channel. The total number of atrial characteristic points is displayed as A _cnt and the total number of ventricular characteristic points is displayed as V _cnt . At 460, a time window is set before each characteristic point of the ventricular channel. At 465, a subtraction count occurs by counting all ventricular characteristic points having at least one atrial characteristic point in a time window in a single pass of a series of ventricular characteristic points. At 470, an addition count occurs by counting all atrial characteristic points that fall within at least one time window associated with a ventricular characteristic point in a single pass of the series of ventricular characteristic points. In one embodiment, these counts can be performed in the same single pass of the ventricular characteristic point. At 475, X and Y values are generated based on the addition count, subtraction count, A _cnt , and V _cnt . In one embodiment, the X value represents a subtraction count expressed as a percentage of the total ventricular characteristic point. In one embodiment, the Y value represents the ratio of the addition count to the subtraction count. In one embodiment, the Y value represents (A: V) ^* V _cnt / A _cnt ² . At 480, the XY points are classified to identify the rhythm type. In one embodiment, the classification of data includes selection of a separation line or separation curve in the XY plane. Data located on one side of the separation line or curve represents a particular rhythm type, and data located on the other side represents a different rhythm type. In one embodiment, the separation line or curve is selected based on a method known as support vector method (SVM) or support vector classification (SVC) (V / Vapnik, “The Nature of Statistical Learning”). Theory ", New York, USA, Springer-Verlag, 1995, et al.). Factors to consider in separating line placement include line placement and slope determination. Considerations include the proximity of the data points and the width of the separation region between different parts of the XY plane. In one embodiment, the XY plane is regionalized in a manner that maximizes the data point separation.

  Other embodiments are also contemplated. For example, in one embodiment, a time window is set after each atrial characteristic point and the count is validated based on the relative timing of the atrial characteristic point and the ventricular characteristic point.

Other criteria are also considered for discrimination between rhythm types. For example, in one embodiment, the identification is based on a percentage of atrial characteristic points that are not related to a ventricular characteristic point or some value derived from a ventricular characteristic point.
Exemplary Device In one embodiment, the present invention includes circuitry, hardware, and software for implantation in the body. In one embodiment, the present invention is adapted for use outside the body.

  FIG. 14 includes a diagram for one embodiment of portions of a heart rate rhythm management system 500 according to the present invention. As illustrated, system 500 includes a device 503, a lead 510, and a lead 540, each coupled to a heart 505. In various embodiments, the device 503 includes an implantable device that functions as a pacemaker, CRT device, cardioverter / defibrillator, pacer / defibrillator, drug delivery device. . In this figure, the device 503 detects cardiac activity from the heart 505 and treats the heart 505 according to the programming and circuitry of the device 503. Leads 510 and 540 include electrodes suitable for electrical connection between the device 503 and the heart 505, each being located at or near the heart 505. Lead 510 is coupled to input circuit 515 and provides an electrical signal sensed from cardiac activity of heart 505. Lead 540 provides treatment to heart 505 in response to the signal received from treatment circuit 535.

  In one embodiment, the device 503 includes an input circuit 515, a controller 520, a telemetry circuit 525, a memory 530, and a therapy circuit 535. Input circuit 515 is coupled to lead 510 and receives an analog signal based on the heart activity of heart 505. In one embodiment, the input circuit 515 includes an analog to digital converter.

  In one embodiment, controller 520 includes a processor and programming for performing the methods described herein. In one embodiment, the controller 520 includes circuitry that controls the therapy circuit 535 and the telemetry circuit 525.

  Memory 530 is coupled to controller 520 and stores data. In one embodiment, the input signal is sampled and characteristic points are generated in near real time. In one embodiment, the characteristic points are generated based on stored sample data derived from the input signal. In one embodiment, programming to be executed by the processor of controller 520 is stored in memory 530.

  In one embodiment, controller 520 uses lead 540 by treatment circuit 535 to determine what treatment or electrical stimulation should be applied to heart 505.

  In one embodiment, controller 520 uses telemetry device 525 to determine what data is to be communicated. In various embodiments, telemetry device 525 is adapted to perform wired or wireless telemetry.

  In one embodiment, the controller 520 performs programming to determine the curvature of the input signal and generate characteristic points.

  In one embodiment, the controller 520 includes a comparator that generates an output based on a comparison of an input signal and a reference signal. A typical comparison includes ascertaining whether the lobe is positively or negatively biased from the baseline. In one embodiment, a typical comparison includes ascertaining whether a series of curved lobes is above or below the deadband threshold. In one embodiment, a typical comparison includes comparing the curvature signal to a threshold hysteresis value.

  In this figure, two leads coupled to the heart 505 are shown. In one embodiment, a single lead provides both sensing and treatment for the heart 505.

  In one embodiment, the system 50 is adapted to determine a heart rate according to the methods described herein. The heart rate is further stored in the memory 530. In one embodiment, the heart rate is used to determine the appropriate treatment or to confirm the delivery of a particular treatment. In one embodiment, the first heart rate is determined by the method described herein, the second heart rate is determined by an auxiliary method, and the controller 520 is configured to detect the first heart rate and the first heart rate. The second heart rate is obtained by collating and adjusting the second heart rate. In one embodiment, the heart rate is communicated by the telemetry circuit 525 to the remote programmer.

In one embodiment, the system 50 is adapted to classify rhythms according to the methods described herein. The rhythm classification is further stored in the memory 530. In one embodiment, classification is used to determine the appropriate treatment or to confirm the delivery of a particular treatment. In one embodiment, the first classification is determined by the method described herein, the second classification for the same time period is determined by the auxiliary method, and the controller 520 is configured with the first classification and the first classification. The second classification is adjusted to determine the third classification. In one embodiment, the classification results are communicated to the remote programmer by telemetry circuit 525.
Alternative Embodiment The characteristic point can be represented by a size corresponding to the area of the lobe in the continuous curvature. In one embodiment, a series of characteristic points are filtered based on their size before calculating heart rate or classifying rhythm types according to the methods described herein. In one embodiment, this requires only selecting characteristic points with a size that exceeds a certain predetermined value that is used to determine heart rate or classify rhythms. In one embodiment, other filtering criteria are applied to the series of characteristic points before determining heart rate or determining rhythm.

  In one embodiment, rhythms are classified based on line segments in the XY plane. In one embodiment, an irregularly shaped region is represented in the xy plane, and a group of data points in the xy plane associated with the region provides an identification criterion. In one embodiment, a probability function is calculated for the data points and serves as an identification criterion.

  In one embodiment, other identification techniques are used in combination with those described herein, and the processor analyzes the differences and makes classification selections.

In one embodiment, the method described herein is performed by a hardware circuit. The circuit can be analog or digital and can include circuit elements such as operational amplifiers or logic gates.
Conclusion The above description is intended to be illustrative and not limiting. Various other embodiments will be apparent to those skilled in the art upon review of the above description.

50 heart rate rhythm management system, 503 device, 510 lead, 515 input circuit, 5
20 controller, 525 telemetry circuit, 530 memory, 535 therapy circuit, 540 leads

Claims (1)

An apparatus as described herein.

Images